February 2012 - armedbear-devel - mailman3.common-lisp.net

[armedbear-devel] Java Collections, ABCL, and you
by Alessio Stalla 04 Feb '12

04 Feb '12

Hello list, I'd like to talk a bit about our Java Collections API integration, and especially a topic that has been lumbering in my mind for quite a lot of time and which has been raised today by Rudi Schlatte on #abcl. Warning: the mail turned out to be quite long. As you probably know, some time ago I added to ABCL an implementation of Christophe Rhodes' extensible sequences proposal[1], until then only implemented in SBCL (as far as I know). It's not loaded by default, but you can get it with (require :extensible-sequences). Some time later I used that API to integrate ABCL with the Java Collections framework, so that (certain) Java collections are Lisp SEQUENCEs. Again, to get it you need to (require :java-collections). However, there are, I think, two interesting limitations in the ext. seq. protocol, which I'm going to discuss. Since it was designed as an extension to CL's sequences, the protocol builds directly upon the concept of CL sequences (lists and vectors) - and in particular it assumes sequences are ordered (accessible by index) and mutable. The only basic operations one must absolutely define to provide a custom sequence implementation are ELT, (SETF ELT), and LENGTH; anything else is defined by default upon those operations, albeit inefficiently. That concept of sequences roughly matches that of Java Lists (except the fact that mutability of Lists is optional). List is a subtype of Collection which implies ordering and access by index. Collection is basically an Iterable thing with a length and optional mutability (in the form of adding elements to the end and removing elements destructively). The concept of a Collection or better an Iterable is missing from the extensible sequences protocol, although it provides an iterator sub-protocol. This makes it unsuitable for those kinds of sequences that do not fully respect CL's definition of sequence (e.g. because they are not mutable of because they can only be iterated once and not accessed by index, like a stream), but that would nevertheless benefit from adhering at least in part to the sequence protocol (because then you might LOOP or, shameless plug, DO+ over them, for example). That would also allow to expose arbitrary Java Collections, and not just Lists, as Lisp sequences - including thus Sets and various specialized Collection types. So the first point is: I'd like to extend the protocol with the concept of an Iterable. How? Ideally, Iterable would be a more basic concept than Sequence, but we cannot modify the class hierarchy of SEQUENCE because it's defined by the standard. So, either ITERABLE must be completely disjoint from SEQUENCE, or it must be a subtype of it. Of the two, I favor the latter, because many sequence operations are supposed to signal errors in case their argument is not a proper sequence, and deviating from that would probably constitute a violation of the standard. So I would define an Iterable as a special SEQUENCE that is defined in terms of an iterator; by default ELT, (SETF ELT) and LENGTH create a new iterator and, very inefficiently, use it to walk the Iterable and do their work. Of course specialized, more efficient implementations are possible. Java Collections would be Iterables, and Java Lists would retain their specialized implementation of ELT and (SETF ELT) which do not necessarily walk the list each time (depending on the type of list). The second idea is that I think it would be beneficial to explicitly state that mutability of sequences is optional. The CL standard doesn't say anything about it (or so it seems to me), although you can clearly see that in certain passages it assumes that list and vector (which are both mutable) are the only subtypes of sequence, whereas list and vector are explicitly said not to constitute an exhaustive partition of the type sequence[2]. Currently a number of sequence functions in the protocol implementation assume mutability, for example SUBSEQ by default creates a copy of the same type and sets elements one by one using an iterator. However, to eliminate the mutability requirement, at least another piece is missing: a generalization of CONS, i.e. an operation that appends a single element to the beginning of a sequence nondestructively. This is roughly what Clojure calls conj (except that conj does not specify where the element is added, which is a very awkward decision to me) and could be defined as (concatenate (class-of sequence) (list item) sequence)... if only concatenate wasn't implemented in terms of (SETF ELT) :P The possible negative sides of such a decision are: 1. it could in the end be an aspect of non-conformity to the ANSI standard (although only relevant to non-standard sequence subtypes) 2. it would make the default implementation of certain sequence operations even more inefficient for custom mutable sequences (that could be mitigated by checking for a marker mixin class MUTABLE-NONSTANDARD-SEQUENCE) 3. it would make ABCL's implementation of sequences a bit more complex. Note also that lifting mutability would NOT benefit Java collections because, inconsistently, they provide no general means to extend a collection without mutating it. What do you think? Cheers, Alessio [1] doc.gold.ac.uk/~mas01cr/papers/ilc2007/sequences-20070301.pd [2] http://www.lispworks.com/documentation/HyperSpec/Body/t_seq.htm

2 1

[armedbear-devel] New fasl format: steps to success
by Erik Huelsmann 04 Feb '12

04 Feb '12

After having mailed a few weeks ago about my ideas regarding a new fasl format, I'm now mailing regarding the next steps that I've come up with to implement the new scheme. But first I'd like to mention that I've come to the conclusion that our "each function is its own class" scheme can't be entirely eliminated: since we don't use invokedynamic yet, there are only two ways to create the function objects that we still require (to stick in the symbol function slot): 1. By using the reflection API to extract the XEPs (eXternal Entry Points) from the fasl class and create a single forwarder class which calls the XEP using reflection or 2. By still creating our multitude of classes which each implement a simple forwarder method which calls the fasl's XEP method I'm not sure of the performance implications of either. Maybe we need to test both schemes before we decide which one would be best. My preference is with method (1), btw, if it performs on par with method (2). So, now about the approach: I. We need to be able to have multiple argument list parsers (the equivalent of Closure.processArgs()) in a single class --> To achieve this, we need to factor out the relevant bits from Closure.java In fact, I did that over the past week: step completed. II. Our compiler should no longer specifically compile output to a class; we need to find a way for the COMPILE and COMPILE-FILE infrastructures to use the same basic compiler which compiles into a context provided by the respective callers. The trick here will be to develop a 'protocol' between the basic compiler and COMPILE/COMPILE-FILE which can be used to flatten the nested structure of compilands. III. Re-enable compilation of the content of the ._ file; we used to have that, but it didn't make a difference in start-up performance, so I disabled it again. Now, we need it to prevent having as many init_xxx() routines as we have functions. After all, at this time, we currently run a mix of ._ and initialization code by intermixing class loading and ._ file execution. IV. Develop a way to split overly big fasl classes, <clinit> (class initialization code) and <init> (constructor code) methods V. Change the base compiler to directly link to method/class combinations when instructed to do so by their caller / through some protocol. VI. COMPILE-FILE should start generating functions and XEPs -- within the current structure VII. Change the file compiler to direct its output to one or a small number of fasl classes, replacing the current output classes with either one of the two options presented earlier in the mail. Step (I) is - as mentioned - already completed. I'm envisioning steps (II) through (VI) to be done on trunk. I expect step (VII) to be associated with experimentation, which would best served by a separate branch. Comments? Ideas? Helpers? :-) Bye, Erik.

1 0

[armedbear-devel] HEADS UP: recent changes require full build
by Erik Huelsmann 04 Feb '12

04 Feb '12

Some of the changes I committed yesterday and today change the ABI of the org.armedbear.lisp.Closure class. Due to this change, a full build is required. Bye, Erik.

1 0

Re: [armedbear-devel] [armedbear-cvs] r13841 - trunk/abcl/contrib/abcl-asdf
by Ville Voutilainen 01 Feb '12

01 Feb '12

On 1 February 2012 15:25, <mevenson(a)common-lisp.net> wrote: > (truename (read-line (sys::process-output > - (sys::run-program "which" `(,mvn-path))))) > + (sys::run-program "which" `(,mvn-path))))) ;; TODO equivalent for MSDOS See http://stackoverflow.com/questions/304319/is-there-an-equivalent-of-which-o…, where.exe does it on windows, where /? gives help for it, and for example C:\Users\vilvouti>where notepad.exe C:\Windows\System32\notepad.exe C:\Windows\notepad.exe So just run 'where' for it and pick the first line if successful, I guess.

2 1

[armedbear-devel] bug report: stable-sort is only stable for lists
by Jorge Tavares 01 Feb '12

01 Feb '12

Dear ABCL developers, Recently I started to investigate the different CL sort implementations and found that stable-sort in ABCL does not execute as required. The problem is that stable-sort is only stable for lists and not for all type of sequences (i.e., elements considered equal by the predicate should stay in their original order). A very simple test shows this error: idesk:~ jast$ abcl Armed Bear Common Lisp 1.0.1 Java 1.6.0_29 Apple Inc. Java HotSpot(TM) 64-Bit Server VM Low-level initialization completed in 0.399 seconds. Startup completed in 1.069 seconds. Loading /Users/jast/.abclrc completed in 4.8 seconds. Type ":help" for a list of available commands. CL-USER(1): (defparameter *stuff* #((1 0.31648433) (0 0.2704757) (0 0.48931926) (1 0.9958766) (0 0.49251676))) *STUFF* CL-USER(2): (setf *stuff* (stable-sort *stuff* #'< :key #'first)) #((0 0.49251676) (0 0.2704757) (0 0.48931926) (1 0.31648433) (1 0.9958766)) The above test is short and simple to understand: *stuff* contains an array where each element is a pair. The first element of a pair is the key and the second the data. With this in mind, the expected result should have been: #((0 0.2704757) (0 0.48931926) (0 0.49251676) (1 0.31648433) (1 0.9958766)). The same error happens with larger arrays and other types of data. I also tested it in SBCL, CCL, ECL and CLisp and for these implementations the results are the expected. Only ABCL differs form the major open source implementations (I didn't test in ACL or LW). The reason for this bug is actually quite simple. In sort.lisp, in the definition of stable-sort, seq-dispatch calls for the non-list sequences the quicksort algorithm, which is not stable. For lists it calls merge sort (which is stable) and the problem does not arise. As a quick fix, I send in attach a patch that uses in stable-sort merge sort for all sequences. This is done by coercing the sequence to list, calling merge sort and coercing it back to the original sequence type. However, as a long term improvement, the best solution would be to implement a merge sort for non-list sequences. Cheers, Jorge Tavares -- http://jorgetavares.com

2 2

[armedbear-devel] Bugs(?) in pathnames
by Theam Yong Chew 01 Feb '12

01 Feb '12

Hi devs, I've observed a few issues (some I'm not so sure about) with pathnames: 1. CL-USER> (make-pathname :device '("c:/a.b.jar") :directory '(:absolute "cl-ppcre")) ; Evaluation aborted on NIL. Why am I not getting any errors (and not dropping into a debugger)? I know the error is because the device is not a list of a pathname ie missing a #P. 2. What's printed out is not what was read in (trailing /) CL-USER> #P"jar:file:c:/a/b.jar!/cl-ppcre/" #P"jar:file:c:/a/b.jar!/cl-ppcre" 3. Should this repeated jar:file:jar:file thing throw an error? #P"jar:file:jar:file:c:/a/b.jar!/cl-ppcre/" ==> #P"jar:file:jar:file:c:/a/b.jar!/cl-ppcre" I think that was all I found, at least for now :-), thanks. Yong.

2 3